Systems configured to identify speech in an audio signal are known. Existing systems, however, typically may rely on an ability to identify phonemes in the signal. A phoneme-based approach may be unreliable at least because phonemes may vary according to context.